Evolutionary Generative Adversarial Networks

By Chaoyue Wang, Chang Xu, Xin Yao, Dacheng Tao. 2018

329 citations recorded by the source corpus.

Generative adversarial networks (GANs) have been effective for learning generative models for real-world data. However, accompanied with the generative tasks becoming more and more challenging, existing GANs (GAN and its variants) tend to suffer from different training problems such as instability and mode collapse. In this paper, we propose a novel GAN framework called evolutionary GANs (E-GANs) for stable GAN training and improved generative performance. Unlike existing GANs, which employ a predefined adversarial objective function alternately training a generator and a discriminator, we evolve a population of generators to play the adversarial game with the discriminator. Different adversarial training objectives are employed as mutation operations and each individual (i.e., generator candidature) are updated based on these mutations. Then, we devise an evaluation mechanism to measure the quality and diversity of generated samples, such that only well-performing generator(s) are preserved and used for further training. In this way, E-GAN overcomes the limitations of an individual adversarial training objective and always preserves the well-performing offspring, contributing to progress in, and the success of GANs. Experiments on several datasets demonstrate that E-GAN achieves convincing generative performance and reduces the training problems inherent in existing GANs.

Subfields: Computer Vision and Pattern Recognition, Statistical and Nonlinear Physics, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Dacheng Tao
2. Cheng Deng
3. Tongliang Liu
4. Fei Wang
5. Chen Qian
6. Chang Xu
7. Andrés Camero
8. Enrique Alba

In Statistical and Nonlinear Physics

1. Tongliang Liu
2. Dacheng Tao
3. Andrés Camero
4. Enrique Alba
5. Changshui Zhang
6. Fei Wang

In Signal Processing

1. Tongliang Liu
2. Shan You
3. Dacheng Tao
4. Andrés Camero
5. Enrique Alba
6. Cheng Deng
7. Chang Xu
8. Changshui Zhang

Citations in the corpus

• GreedyNAS: Towards Fast One-Shot NAS With Greedy Supernet (2020, background): “By dint of automatic feature engineering, deep neural networks (DNNs) have achieved remarkable success in various computer vision tasks, such as image classification [37, 36, 41, 33, 32, 15, 39], visual generation [34, 35], image retrieval [40, 7, 8, 12] and semantic comprehension [18, 17].”
• Perceptual Adversarial Networks for Image-to-Image Transformation (2017, background): “Some of these GANs-based works are committed to training a better generative model, such as InfoGAN [37], Energy-based GAN [38], WGAN(-GP) [39], [40], Progressive GAN [41], E-GAN [42] and SN-GAN [43].”
• Modeling Image Composition for Complex Scene Generation (2022, background): “For better controllability, LostGANs [32, 33] first synthesize the semantic masks from layouts, and then ISLA-Norm is proposed for generating color images from specific masks and style codes.”
• Bio-inspired computation: Where we stand and what's next (2019, unknown): “[434] evolved a population of generators which”
• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications (2020, unknown): “Evolutionary GANs are proposed in [288].”
• Unsupervised Semantic-Preserving Adversarial Hashing for Image Search (2019, methodology): “[58], have been widely used in computer vision to generate realistic images [59], [60].”
• A Comprehensive Overview and Comparative Analysis on Deep Learning Models: CNN, RNN, LSTM, GRU (2023, background): “The generator's updating gradients are determined by the discriminator through an adaptive objective [65].”
• Random error sampling-based recurrent neural network architecture optimization (2019, methodology): “Evolutionary GAN (E-GAN) applies ES to evolve a population of networks (generators) that are mutated by applying SGD according to di ff erent loss functions [27].”